東華大學圖書館 |

Novel Decomposition Methods to Enable Interpretable Artificial Intelligence and Scheduling /

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Novel Decomposition Methods to Enable Interpretable Artificial Intelligence and Scheduling // Tengmu Hu.
作者:	Hu, Tengmu,
面頁冊數:	1 electronic resource (193 pages)
附註:	Source: Dissertations Abstracts International, Volume: 86-11, Section: B.
Contained By:	Dissertations Abstracts International86-11B.
標題:	Industrial engineering. -
電子資源:	https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=32111744
ISBN:	9798314889510

Novel Decomposition Methods to Enable Interpretable Artificial Intelligence and Scheduling /
Hu, Tengmu,

Novel Decomposition Methods to Enable Interpretable Artificial Intelligence and Scheduling /Tengmu Hu. - 1 electronic resource (193 pages)

Source: Dissertations Abstracts International, Volume: 86-11, Section: B.

Integer programming relates to optimization problems that allow people to make better quality decisions faster. The word "integer" means that some of the decisions involve integer variables such as which variable is included, or which part is made. Here, we focus on two main types of applications. The first relates to so-called "interpretable artificial intelligence" and classifying automatically observations in a way that people can understand. The technical name for this work is "Optimal Classification Trees" (OCTs) and we seek methods to generate these trees more quickly and at high quality than previous methods. The second problem relates to scheduling operations with sequence dependent setups. More specifically, we focus on scheduling automotive stamping operations which make many of the parts which become parts of the cars including bodies and power trains.Optimal Classification Tree (OCT) formulations combine interpretability with predictive power, but recent methods face challenges like excessive decision variables, invalid tree structures, and inefficiencies on large datasets. Our contributions of OCT are as follows:Modified OCT(M-OCT): Introduces novel leaf-branch interaction constraints to stabilize tree structure and reduce binary variables via threshold encoding. Tested on 15 datasets and 750 cases, M-OCT improves accuracy, efficiency, and structure. It achieves 94% accuracy in defect detection for pipe sections.Mixed Set-packing and Set-partition Formulation with Column Generation: We develop a mixed set-packing and set-partition formulation and apply a column generation approach using binary-encoded thresholds and dynamic programming to solve subproblems. The Integer Column Generation (ICG) algorithm we propose eliminates the need for commercial solvers and achieves 97% accuracy in predicting cyber vulnerabilities across 10 datasets.OCT with Rejection Option: To enhance interpretability, we propose a rejection option that increases leaf node purity. This model outperforms standard OCT in the number of pure leaves and correct predictions.Scheduling problems with sequence dependent setups are well-known NP-hard problems. In the sheet metal stamping manufacturing process, products belong to different families, and switching between families incurs a setup time. This makes deciding which family to use and how to produce the products more complex. For manufacturing companies that produce steel bars, machines can only process raw materials of specific sizes. Changing the size requires additional setup time.Wedefinethesheetmetalstampingmanufacturingprocessasasingle-machinescheduling problem with sequence dependent setups where switching between product families incurs costly setup times. For these types of problems, solution quality is crucial, as frequent machine setups can create challenges for onsite workers, and delayed products can harm a company's reputation. Existing state-of-the-art methods using meta-heuristics struggle to guarantee solution quality, while exact methods often face computational efficiency issues. To address these challenges, we propose a novel decomposition method specifically designed for such problems. Our proposed three-stage algorithmic framework includes:Efficient intra-family scheduling via a novel heuristic with complexity O(n5logn) where n is the number of jobs. Dynamic programming for family sequence optimization. •A branch-and-bound method (B&B-DP) to optimize family-switch timing.This algorithmic framework extends the capabilities of single-machine scheduling with family setup times to handle a large number of jobs. Our proposed branch-and bound-regulated dynamic programming(B&B-DP) algorithm excels in solving large-scale scheduling problems, demonstrating superior performance against four benchmark methods across 150 test cases. In our experiments, we show that the proposed algorithm reduces total tardiness by 10%-25% compared to other methods.

English

ISBN: 9798314889510Subjects--Topical Terms:

526216
Industrial engineering.
Subjects--Index Terms:

Optimal classification tree

Novel Decomposition Methods to Enable Interpretable Artificial Intelligence and Scheduling /
LDR:05363nmm a22004213i 4500 001 2432734
005 20250731101922.5
006 m o d
007 cr|nu||||||||
008 261011s2025 miu||||||m |||||||eng d
020 $a 9798314889510
035 $a (MiAaPQD)AAI32111744
035 $a (MiAaPQD)OhioLINKosu1744905662294894
035 $a AAI32111744
040 $a MiAaPQD $b eng $c MiAaPQD $e rda
100 1 $a Hu, Tengmu, $e author. $3 3822072
245 1 0 $a Novel Decomposition Methods to Enable Interpretable Artificial Intelligence and Scheduling / $c Tengmu Hu.
264 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2025
300 $a 1 electronic resource (193 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 86-11, Section: B.
500 $a Advisors: Allen, Theodore.
502 $b Ph.D. $c The Ohio State University $d 2025.
520 $a Integer programming relates to optimization problems that allow people to make better quality decisions faster. The word "integer" means that some of the decisions involve integer variables such as which variable is included, or which part is made. Here, we focus on two main types of applications. The first relates to so-called "interpretable artificial intelligence" and classifying automatically observations in a way that people can understand. The technical name for this work is "Optimal Classification Trees" (OCTs) and we seek methods to generate these trees more quickly and at high quality than previous methods. The second problem relates to scheduling operations with sequence dependent setups. More specifically, we focus on scheduling automotive stamping operations which make many of the parts which become parts of the cars including bodies and power trains.Optimal Classification Tree (OCT) formulations combine interpretability with predictive power, but recent methods face challenges like excessive decision variables, invalid tree structures, and inefficiencies on large datasets. Our contributions of OCT are as follows:Modified OCT(M-OCT): Introduces novel leaf-branch interaction constraints to stabilize tree structure and reduce binary variables via threshold encoding. Tested on 15 datasets and 750 cases, M-OCT improves accuracy, efficiency, and structure. It achieves 94% accuracy in defect detection for pipe sections.Mixed Set-packing and Set-partition Formulation with Column Generation: We develop a mixed set-packing and set-partition formulation and apply a column generation approach using binary-encoded thresholds and dynamic programming to solve subproblems. The Integer Column Generation (ICG) algorithm we propose eliminates the need for commercial solvers and achieves 97% accuracy in predicting cyber vulnerabilities across 10 datasets.OCT with Rejection Option: To enhance interpretability, we propose a rejection option that increases leaf node purity. This model outperforms standard OCT in the number of pure leaves and correct predictions.Scheduling problems with sequence dependent setups are well-known NP-hard problems. In the sheet metal stamping manufacturing process, products belong to different families, and switching between families incurs a setup time. This makes deciding which family to use and how to produce the products more complex. For manufacturing companies that produce steel bars, machines can only process raw materials of specific sizes. Changing the size requires additional setup time.Wedefinethesheetmetalstampingmanufacturingprocessasasingle-machinescheduling problem with sequence dependent setups where switching between product families incurs costly setup times. For these types of problems, solution quality is crucial, as frequent machine setups can create challenges for onsite workers, and delayed products can harm a company's reputation. Existing state-of-the-art methods using meta-heuristics struggle to guarantee solution quality, while exact methods often face computational efficiency issues. To address these challenges, we propose a novel decomposition method specifically designed for such problems. Our proposed three-stage algorithmic framework includes:Efficient intra-family scheduling via a novel heuristic with complexity O(n5logn) where n is the number of jobs. Dynamic programming for family sequence optimization. •A branch-and-bound method (B&B-DP) to optimize family-switch timing.This algorithmic framework extends the capabilities of single-machine scheduling with family setup times to handle a large number of jobs. Our proposed branch-and bound-regulated dynamic programming(B&B-DP) algorithm excels in solving large-scale scheduling problems, demonstrating superior performance against four benchmark methods across 150 test cases. In our experiments, we show that the proposed algorithm reduces total tardiness by 10%-25% compared to other methods.
546 $a English
590 $a School code: 0168
650 4 $a Industrial engineering. $3 526216
650 4 $a Systems science. $3 3168411
653 $a Optimal classification tree
653 $a Integer programming
653 $a Dynamic programming
653 $a Scheduling
690 $a 0546
690 $a 0790
710 2 $a The Ohio State University. $b Industrial and Systems Engineering. $e degree granting institution. $3 3816541
720 1 $a Allen, Theodore $e degree supervisor.
773 0 $t Dissertations Abstracts International $g 86-11B.
790 $a 0168
791 $a Ph.D.
792 $a 2025
856 4 0 $u https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=32111744